MON-77 Azure Event Hub monitors

This commit is contained in:
Laurent Piroelle 2017-10-30 11:40:16 +01:00
parent 22e7021681
commit d2e1aa5efd
3 changed files with 128 additions and 0 deletions

View File

@ -0,0 +1,31 @@
variable "environment" {}
variable "down_message" {}
variable "failed_requests_message" {}
variable "errors_message" {}
variable "delay" {
default = 600
}
variable "failed_requests_rate_thresold_critical" {
default = 5
}
variable "failed_requests_rate_thresold_warning" {
default = 3
}
variable "errors_rate_thresold_critical" {
default = 5
}
variable "errors_rate_thresold_warning" {
default = 3
}
variable "use_filter_tags" {
default = "true"
}

View File

@ -0,0 +1,86 @@
resource "datadog_monitor" "eventhub_status" {
name = "[${var.environment}] Event Hub status"
message = "${var.down_message}"
query = <<EOF
avg(last_5m): avg:azure.eventhub_namespaces.status{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} != 1
EOF
type = "query alert"
notify_no_data = true
evaluation_delay = "${var.delay}"
renotify_interval = 60
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
no_data_timeframe = 20
}
resource "datadog_monitor" "eventhub_failed_requests" {
name = "[${var.environment}] Event Hub failed requests"
message = "${var.failed_requests_message}"
query = <<EOF
avg(last_5m): (
avg:azure.eventhub_namespaces.failed_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
) * 100 / (
avg:azure.eventhub_namespaces.successful_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
avg:azure.eventhub_namespaces.failed_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
) > ${var.failed_requests_rate_thresold_critical}
EOF
type = "query alert"
thresholds {
critical = "${var.failed_requests_rate_thresold_critical}"
warning = "${var.failed_requests_rate_thresold_warning}"
}
notify_no_data = false
evaluation_delay = "${var.delay}"
renotify_interval = 60
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
no_data_timeframe = 20
}
resource "datadog_monitor" "eventhub_errors" {
name = "[${var.environment}] Event Hub errors"
message = "${var.errors_message}"
query = <<EOF
avg(last_5m): (
avg:azure.eventhub_namespaces.internal_server_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
avg:azure.eventhub_namespaces.server_busy_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
avg:azure.eventhub_namespaces.other_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
) * 100 / (
avg:azure.eventhub_namespaces.successful_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
avg:azure.eventhub_namespaces.internal_server_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
avg:azure.eventhub_namespaces.server_busy_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
avg:azure.eventhub_namespaces.other_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
) > ${var.errors_rate_thresold_critical}
EOF
type = "query alert"
thresholds {
critical = "${var.errors_rate_thresold_critical}"
warning = "${var.errors_rate_thresold_warning}"
}
notify_no_data = false
evaluation_delay = "${var.delay}"
renotify_interval = 60
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
no_data_timeframe = 20
}

View File

@ -0,0 +1,11 @@
output "status_monitor_id" {
value = "${datadog_monitor.eventhub_failed_requests.id}"
}
output "failed_requests_monitor_id" {
value = "${datadog_monitor.eventhub_status.id}"
}
output "errors_monitor_id" {
value = "${datadog_monitor.eventhub_errors.id}"
}